Context Trails

Repository containing the code used in the paper:
Context Trails: A Dataset to Study Contextual and Tour Recommendation

📌 Requirements

To run the code in this repository, you need the following prerequisites:

🖥️ Required Software

• Java (version 8 or higher)
• Mono (to run MyMediaLite)
• Matlab (to run IRenMF)
• Python (version 3.8 or higher)

📦 Python Dependencies

To install the required libraries, run the following command:

pip install \
    Cython==0.29.6 \
    h5py==2.9.0 \
    nltk==3.4.5 \
    nose==1.3.7 \
    numpy==1.16.2 \
    pandas==0.24.2 \
    scikit-learn==0.20.3 \
    scikit-optimize==0.5.2 \
    scipy==1.2.1 \
    tqdm==4.31.1 \
    implicit==0.4.4 \
    similaripy==0.1.2

Probably, you will need to install also these libraries:

  sudo apt install g++
  conda install -c conda-forge libstdcxx-ng

📂 Repository Structure

• online_appendix.pdf: file with extended tables from those shown in the paper.
• src: directory containing the source code of the recommenders used in the experiments
• scripts: directory containing the bash scripts for generating the recommendations. IMPORTANT: ensure that scripts end with the end-of-line character of LF, not CRLF.

• POI: contains the scripts to perform POI recommendations. Follow this order of execution:
  • step1_prepare_data.sh: script that will prepare all necessary files to execute the experiments for POI recommendation. NOTE: configure properly the ''path_source_data_repository'' variable. This variable must reference to the path of the zenodo repository Context Trails data.
  • step2_generate_recommenders.sh: script that will execute the experiments for POI recommendation. It will generates a recommendation folder and a results folder for each city. The format of each recommendation file (one per each hyperparameter configuration of each recommender) is: user_id \t venue_id \t score. The format of each result file (one per each hyperparameter configuration of each recommender) is: metric \t result. NOTE:you will need to set the ''fullpathMatlab'' variable to refer to the path of your Matlab installation.
• Route: contains the scripts to perform route recommendations. Order of execution:
  • step1_prepare_routes_data.sh: script that will prepare all necessary files to execute the experiments for Route recommendation. NOTE: configure properly the ''path_source_data_repository'' variable. This variable must reference to the path of the zenodo repository Context Trails data.
  • step2_generate_route_recommenders.sh: script that will execute the experiments for Route recommendation. It will generates a recommendation folder and a results folder for each city. Each recommendation/result file will follow the same format as in POI recommendation.

• NOTE: the data used in this repository can be found in the zenodo repository Context Trails data. It is necessary to have the data repository in order to run the experiments correctly.

🛠️ Additional algorithms/frameworks used in the experiments

• MyMedialite (for BPR)
• Embarrassingly Shallow Autoencoders for Sparse Data (for EASEr): source code (directory ease_rec).
• Updatable, Accurate, Diverse, and Scalable Recommendations for Interactive Applications (for RP3beta): source code (directory recommender-systems).

👥Authors

• Pablo Sánchez - Universidad Pontificia Comillas
• Alejandro Bellogín - Universidad Autónoma de Madrid
• José Luis Jorro Aragoneses - Universidad Autónoma de Madrid

📬 Contact

• Pablo Sánchez - psperez@icai.comillas.edu

🙏 Acknowledgments

• This work was supported by grant PID2022-139131NB-I00 funded by MCIN/AEI/10.13039/501100011033 and by ``ERDF A way of making Europe.''